|Home | About | Journals | Submit | Contact Us | Français|
A midbrain network interacts with the well-known frontoparietal forebrain network to select stimuli for gaze and spatial attention. The midbrain network, containing the superior colliculus (SC; optic tectum, OT, in non-mammalian vertebrates) and the isthmic nuclei, helps evaluate the relative priorities of competing stimuli and encodes them in a topographic map of space. Behavioral experiments in monkeys demonstrate an essential contribution of the SC to stimulus selection when the relative priorities of competing stimuli are similar. Neurophysiological results from the owl OT demonstrate a neural correlate of this essential contribution of the SC/OT. The multi-layered, spatiotopic organization of the midbrain network lends itself to the analysis and modeling of the mechanisms underlying stimulus selection for gaze and spatial attention.
Animals are subjected to a constant and potentially overwhelming barrage of information from the environment. Their survival depends on the ability to correctly identify and process the most important information at every instant in time. The neural computations thought to achieve this goal center around three major steps: the evaluation of relative stimulus importance (“priority”), the selection of the location in the environment with the highest priority, and the deployment of attention or action to that location (Figure 1) [1-3]. These computational steps are carried out in overlapping populations of neurons .
The midbrain and the forebrain contain networks involved in performing these computations . The midbrain network contains the superior colliculus (SC; optic tectum, OT, in non-mammalian vertebrates) and the isthmic nuclei. Working together with the frontoparietal network in the forebrain [6-8], it evaluates the relative priorities of stimuli and maintains a topographic representation of those priorities [1,3,9]. This representation is updated rapidly, in real-time, as stimuli and the animal’s behavioral state change and is transmitted to other brain regions that control attention, analyze stimulus features, signal important events for reinforcement, and execute orienting movements [9-13].
Although both the midbrain and the forebrain networks are capable of carrying out most of these computational steps, recent evidence indicates that the midbrain network (specifically, the intermediate and deep layers of the SC) is necessary for selecting the highest priority stimulus when an animal encounters multiple stimuli of similar priorities. This essential role of the midbrain network in selecting the highest priority stimulus under difficult conditions adds to its well-established roles in multimodal sensorimotor integration, salience and priority evaluation, and gaze control [1,3,14,15].
The necessity of the SC for stimulus selection was demonstrated in monkeys performing a color or contrast oddball search task (Figure 2A) . Focal inactivation of the SC caused the monkey to select any of the four stimuli, rather than just the oddball stimulus, when the oddball stimulus was located in the inactivated portion of the SC space map. The impairment was specific to selection among multiple, competing stimuli and not due to visual or motor deficits. When the difference between the oddball and the other stimuli was large (easy task), the impairment resulting from SC inactivation was small (Figure 3A). However, when the difference among the stimuli was small (hard task) the impairment was profound (Figure 3A).
A similar effect of SC inactivation was demonstrated in monkeys performing a cued spatial attention task (Figure 2B) . Focal SC inactivation caused the monkey to report the direction of the task-relevant distracter, rather than the direction of the cued stimulus, when the cued stimulus was located in the inactivated portion of the SC space map. The impairment was specific to selection among similar, competing stimuli and could not be explained by visual or motor deficits. When the difficulty of the task was reduced, by replacing the task-relevant distracter with a task-irrelevant distracter, SC inactivation resulted in almost no impairment of performance. Thus, as in the previous experiment, SC inactivation affects stimulus selection most when the difference among stimuli is small, while having little or no effect when the difference is large.
Other studies confirm the necessity of the SC in different types of stimulus selection tasks (monkeys: ; rats performing an odor discrimination task: ). The relationship between task difficulty and performance impairment following SC inactivation was observed in these tasks as well (but see ). In addition, focal electrical microstimulation applied to the SC results in a systematic improvement in the selection or discrimination of stimuli located in the activated portion of the SC space map, demonstrating the SC’s causal role in competitive stimulus selection [21-24].
In summary, these results reveal that an essential role of the SC is specific to the selection of one among many competing stimuli and that the importance of the SC increases with increasing difficulty of the selection task. We next describe neural correlates of the computational steps (Figure 1) that lead to the selection of the highest priority stimulus by the midbrain network.
The identification of the highest priority stimulus begins with the evaluation of the priorities of stimuli (Figure 1). The key factors that contribute to evaluating the priority of a stimulus are its physical salience and its behavioral relevance. The SC/OT receives information about both of these factors. Most neurons in the SC/OT encode the physical salience (strength) of stimuli in their classical receptive fields (RFs). They respond vigorously to novel stimuli  and with increasing firing rates to increasing intensity or motion speed (strength) of the stimulus [26-29]. On the other hand, they exhibit little selectivity for stimulus features, such as color, orientation, motion direction, or sound frequency [17,26,30-34].
The behavioral relevance of a stimulus depends primarily on three factors (Figure 1). One factor is whether an animal associates the stimulus with rewarding or aversive consequences as a result of learning or innate predisposition. The influence of behavioral relevance in modulating SC/OT responses has been demonstrated by the finding that stimulus feature tuning can develop in SC neurons when the stimulus is associated with a reward ; reward conditioning dramatically increases SC/OT responses to the stimulus. Indeed, the strength of responses can be modulated in real-time by the size or probability of the reward [36,37]. Other factors that contribute to the representation of behavioral relevance in the SC are the selection of a stimulus for attention [38-40] or motor plans to orient gaze towards the location of the stimulus [41-44]. All three factors can increase SC/OT responses to a stimulus. In the midbrain network, the priorities of stimuli are encoded in terms of spike rates in a map of space, with higher firing rates corresponding to higher stimulus priority.
The mechanism that computes relative stimulus priority is the global competitive surround. Studies in mammals and birds demonstrate that stimuli of any sensory modality, located far outside of the classical RF, can suppress the responses of SC/OT neurons to stimuli located inside their RFs [45-49]. A recent study of the owl OT characterized the spatial dependence of these competitive interactions in detail . Another study explored the rules of relative priority representation in the OT . Neuronal responses in the owl OT were recorded while systematically changing the relative priorities of competing stimuli; achieved by varying their relative strengths (salience) but not their behavioral relevance (Figure 2C) . The responses of most OT neurons were suppressed by the presence of the competing stimulus. About half of these neurons showed a gradual increase in response suppression as the strength of the competitor was increased, referred to as “gradual” suppression (Figure 3C). When these neurons were retested with a stronger stimulus inside the RF, the same gradual suppression was observed along with an overall increase in the neuronal firing rates. The results demonstrate that firing rates in the OT represent both the relative and the absolute stimulus strengths (priorities) of stimuli. The spatial map of relative stimulus priorities forms the basis for stimulus selection.
Selection of the highest priority stimulus involves an operation (a “decoder”) that compares firing rates across spatial locations in the relative priority map and determines the peak in the distribution of firing rates. A winner-take-all operation is one such decoder. It selects the location with the highest firing rate regardless of its difference from the next highest firing rate . In contrast, the ideal observer analysis [3,51] is a probabilistic decoder. It selects the location with the highest firing rate with a probability that reflects how well the distributions of the highest and the next highest firing rates are separated. How the brain actually implements peak selection is not known. However, a recent study has shown that the probabilistic method correctly predicted the behavioral performance of monkeys from SC firing rates in a selection task on more trials than the winner-take-all method, with both methods performing equally well when the rate distributions were well separated .
The probabilistic method has been applied in several recent studies to decode SC/OT firing rates [27,28,35,53]. These studies have shown that the discriminability of the peak rate (probability of selection) decreases as the difference between the priorities of the competing stimuli decreases (for instance, Figure 3B, data in black; ). The behavioral observation that the normal decrease in selection performance with decreasing stimulus differences becomes far more severe following SC inactivation (Figure 3A, [16-19]) suggests that the SC/OT provides an additional computational step that improves stimulus discriminability prior to stimulus selection (Figure 1, box in green), as described next.
We propose that one of the essential computational contributions of the midbrain network to stimulus selection is to improve the discriminability of the peak in the relative stimulus priority map in the SC/OT, specifically when the priorities of the competing stimuli are similar.
Neural correlates of such an essential computation (Figure 3B) were reported in the recent study that measured OT responses in owls to pairs of simultaneous stimuli (discussed previously, Figure 2C, ). Although many OT neurons exhibited gradual response suppression by a competing stimulus of increasing strength (Figure 3C), about 30% of the neurons exhibited “switch-like” suppression (Figure 3D): responses changed abruptly from a high level to a lower level as the strength of the competing stimulus increased, and this abrupt transition occurred when the strength of the competitor just exceeded that of the RF stimulus. When compared to gradual responses, switch-like responses improved the discriminability of the peak in the relative priority map specifically when the competing stimuli were similar (Figure 3B, data in green). This improvement resulted from an increase in firing rates to the strongest (highest priority) stimulus and a decrease in firing rates to the next strongest stimulus (Figure 3E; ). This computational step gave rise to a population-wide categorical representation of the strongest stimulus . Interestingly, when a single stimulus was presented alone, there was no improvement in peak discriminability in the responses of switch-like versus gradual neurons [27,54], consistent with the little to no effect of SC inactivation on the selection of single stimuli. Thus, improved peak discriminability of the relative priority map is a neural correlate of an essential contribution of the midbrain network to competitive stimulus selection.
We next explore mechanisms that could underlie the construction of the relative priority map in the SC/OT and the improvement in its peak discriminability.
Globally projecting circuits, intrinsic to the SC/OT, that might support the generation of a relative priority map in the SC/OT ([27,50]) have been sought, but not found . However, as recognized first by Sereno and Ulinski when studying the turtle midbrain , the unusual anatomy of the isthmic nuclei is perfectly suited to serve this function. The isthmic nuclei, located beneath the SC/OT in the lateral tegmentum, contain two major circuits. One circuit consists of a population of cholinergic neurons that are connected reciprocally and topographically with the SC/OT. In birds, these neurons are in two adjacent nuclei called the nucleus isthmi pars parvocellularis (Ipc) and the nucleus semilunaris (SLu) (Figure 4, blue ); in mammals, these neurons cluster in a single nucleus called the parabigeminal nucleus [58,59]. A second circuit, most thoroughly described in birds, consists of a population of GABAergic neurons that receive topographic input from the SC/OT and send projections back broadly to the space map in the SC/OT and to the cholinergic nuclei (Figure 4, red ). In birds, these neurons are in the nucleus isthmi pars magnocellularis (Imc); in mammals, the analogous neurons are thought to be in the lateral tegmental nucleus [59,61,62]
The cholinergic isthmic circuit is thought to provide space-specific, feedback amplification of SC/OT responses, although this function has yet to be demonstrated directly. In owls, Ipc neurons are multimodal, sharply tuned for space, and respond according to the physical salience of stimuli . In cats, the responses of PBN neurons are enhanced by motor plans for orienting eye movements . Importantly, the responses of many Ipc neurons in owls are suppressed in a switch-like manner with increasing strength of a distant competing stimulus , suggesting that Ipc activity contributes, at least in part, to switch-like responses in the OT.
The GABAegic isthmic circuit is thought to mediate global competition in the SC/OT, although this function has yet to be demonstrated directly. In pigeons, Imc neurons are responsible for long-range inhibition in the Ipc . Thus, the Imc is likely to mediate global competition in the OT both by regulating feedback amplification provided by the cholinergic isthmic circuit and by directly inhibiting OT neurons.
It has been proposed by many authors that the SC/OT, together with the isthmic circuits, form a module for the implementation of a winner-take-all evaluation of the most salient stimulus [56,66]. A detailed model of this module accounted for some, but not all, of the key aspects of switch-like responses in the OT , leaving open the question of the computational rules that implement improved peak discrimination in the midbrain network.
The role of the midbrain network in multimodal, sensorimotor integration and gaze control are well established. Its role in controlling attention is just being fully appreciated. One essential role of this network is to enable an animal to identify the highest priority stimulus among multiple stimuli with similar priorities. Switch-like response suppression when a competing stimulus becomes the strongest stimulus, as observed in the OT and in the cholinergic isthmic circuit in owls, provides a compelling neural correlate for this essential role.
It will be important to test whether switch-like response suppression and improved peak discriminability occur in the SC/OT of other species as well, particularly in monkeys. A behavioral protocol that will facilitate the observation of these phenomena is one that includes both the systematic variation of relative stimulus priorities of two or more simultaneous, task-relevant stimuli and a delayed response task (to allow for sufficient time for the neural feedback mechanisms of the isthmic circuits to exert their influence on SC/OT responses). In addition, further experimental and modeling investigations will be needed to understand the roles of the cholinergic and GABAergic isthmic circuits, both in the construction of the relative priority map and in improving peak discriminability in the OT. Equally important, the functional properties of the analogous circuits in mammals need to be explored in far greater detail.
Funding from the Stanford University School of Medicine (Dean’s Postdoctoral Fellowship, SPM) and the NIH (R01 EY019179 to EIK) is gratefully acknowledged. We thank Phyllis Knudsen for preparing the figures. We also thank Ali Asadollahi, Astra Bryant, Cathy Dunn, Yvette Fisher, Alex Goddard, and Jason Schwarz for critically reading the manuscript.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
• of special interest
of outstanding interest